Guiding Principle for Minor-Embedding in Simulated-Annealing-Based Ising Machines

TL;DR

This paper introduces a new guiding principle for minor-embedding in simulated-annealing-based Ising machines, improving their ability to solve complex combinatorial optimization problems efficiently.

Contribution

It proposes a statistically-mechanics-based minor-embedding method that outperforms existing methods in simulated annealing for Ising models.

Findings

Proposed ME outperforms existing MEs across benchmarks.

Performance improves notably with high degree variance in logical models.

Theoretical basis enhances embedding efficiency and solution quality.

Abstract

We propose a novel type of minor-embedding (ME) in simulated-annealing-based Ising machines. The Ising machines can solve combinatorial optimization problems. Many combinatorial optimization problems are mapped to find the ground (lowest-energy) state of the logical Ising model. When connectivity is restricted on Ising machines, ME is required for mapping from the logical Ising model to a physical Ising model, which corresponds to a specific Ising machine. Herein we discuss the guiding principle of ME design to achieve a high performance in Ising machines. We derive the proposed ME based on a theoretical argument of statistical mechanics. The performance of the proposed ME is compared with two existing types of MEs for different benchmarking problems. Simulated annealing shows that the proposed ME outperforms existing MEs for all benchmarking problems, especially when the distribution of the degree in a logical Ising model has a large standard deviation. This study validates the guiding principle of using statistical mechanics for ME to realize fast and high-precision solvers for combinatorial optimization problems.